Repository navigation
Conversation
Fixes ggml-org#11 This fixes a Japanese prompt I was attempting to run EG: `./main -m ./models/13B/ggml-model-q4_0.bin -t 8 -n 128 -n 512 -p $'人生の意味は'` Output before change: `人生の意���、フロントカードに���いてる。 2019年3月 © All Rights Reserved. [end of text]` So it is outputting some characters but some � Output after change: `人生の意は、一人が一人ということであります。は安部が立していたので、去からは一人の人にれるのはにとどまったのですが、そう`
Contributor
Author
|
Closing the PR because what is really needed is different tokenization mechanism see discussion here: |
Hades32
pushed a commit
to Hades32/llama.cpp
that referenced
this pull request
Mar 21, 2023
add easy Windows install instructions to the readme
flowgrad
pushed a commit
to flowgrad/llama.cpp
that referenced
this pull request
Jun 27, 2023
fix bug: Parameter --reverse-prompt won't accept text
4 tasks done
jesusmb1995
pushed a commit
to jesusmb1995/llama.cpp
that referenced
this pull request
Oct 30, 2025
rururush
pushed a commit
to USTC-ADSL/llama.cpp
that referenced
this pull request
Mar 16, 2026
* fix warning * wip * add todo for graph key generate * rename some file to meet upstream guideline * remove local .clang-format * expend supported/unsupported counter to all ops * append device name to log * port to ggml logger * fix warning after adapt to ggml logger * append \n to all log * use case op instead of convert * Revert "use case op instead of convert" This reverts commit e662fc2. * fix op that needs same shape * opt kQnnOpsTable * refresh params name field when getting op config * opt npu log print * remove unused functions
TheTom
referenced
this pull request
in TheTom/llama-cpp-turboquant
Mar 25, 2026
Massive reduction in constant memory and compute: - 256KB of dense matrices → 512 bytes of sign arrays - O(d²) = 16,384 ops → O(d log d) = 896 ops per rotation - Metal shader file: 1.5MB → 432KB Speed: still 2.4 tok/s. WHT reduced per-rotation cost but the bottleneck is redundant calls (8-32× per block from flash attention). The dequantize function is called per 4/16-element chunk, each time doing the full 128-element WHT. Need to modify the flash attention kernel to dequantize once per block. Quality: WHT+signs gives BETTER quality than dense QR on real KV tensors (cosine 0.94 vs 0.79 at 2-bit). Sub-Gaussian distribution (kurtosis 1.53) means fewer outliers hitting extreme centroids. Reviewed by Codex: WHT butterfly correct, inverse order verified, QJL correction matches reference C implementation. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
referenced
this pull request
in TheTom/llama-cpp-turboquant
Mar 26, 2026
Massive reduction in constant memory and compute: - 256KB of dense matrices → 512 bytes of sign arrays - O(d²) = 16,384 ops → O(d log d) = 896 ops per rotation - Metal shader file: 1.5MB → 432KB Speed: still 2.4 tok/s. WHT reduced per-rotation cost but the bottleneck is redundant calls (8-32× per block from flash attention). The dequantize function is called per 4/16-element chunk, each time doing the full 128-element WHT. Need to modify the flash attention kernel to dequantize once per block. Quality: WHT+signs gives BETTER quality than dense QR on real KV tensors (cosine 0.94 vs 0.79 at 2-bit). Sub-Gaussian distribution (kurtosis 1.53) means fewer outliers hitting extreme centroids. Reviewed by Codex: WHT butterfly correct, inverse order verified, QJL correction matches reference C implementation. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
didlawowo
pushed a commit
to didlawowo/llama.cpp
that referenced
this pull request
Mar 27, 2026
…gml-org#26 Massive reduction in constant memory and compute: - 256KB of dense matrices → 512 bytes of sign arrays - O(d²) = 16,384 ops → O(d log d) = 896 ops per rotation - Metal shader file: 1.5MB → 432KB Speed: still 2.4 tok/s. WHT reduced per-rotation cost but the bottleneck is redundant calls (8-32× per block from flash attention). The dequantize function is called per 4/16-element chunk, each time doing the full 128-element WHT. Need to modify the flash attention kernel to dequantize once per block. Quality: WHT+signs gives BETTER quality than dense QR on real KV tensors (cosine 0.94 vs 0.79 at 2-bit). Sub-Gaussian distribution (kurtosis 1.53) means fewer outliers hitting extreme centroids. Reviewed by Codex: WHT butterfly correct, inverse order verified, QJL correction matches reference C implementation. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
InfernalDread
referenced
this pull request
in InfernalDread/llama.cpp
Apr 4, 2026
Massive reduction in constant memory and compute: - 256KB of dense matrices → 512 bytes of sign arrays - O(d²) = 16,384 ops → O(d log d) = 896 ops per rotation - Metal shader file: 1.5MB → 432KB Speed: still 2.4 tok/s. WHT reduced per-rotation cost but the bottleneck is redundant calls (8-32× per block from flash attention). The dequantize function is called per 4/16-element chunk, each time doing the full 128-element WHT. Need to modify the flash attention kernel to dequantize once per block. Quality: WHT+signs gives BETTER quality than dense QR on real KV tensors (cosine 0.94 vs 0.79 at 2-bit). Sub-Gaussian distribution (kurtosis 1.53) means fewer outliers hitting extreme centroids. Reviewed by Codex: WHT butterfly correct, inverse order verified, QJL correction matches reference C implementation. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
didlawowo
pushed a commit
to didlawowo/llama.cpp
that referenced
this pull request
Apr 4, 2026
…gml-org#26 Massive reduction in constant memory and compute: - 256KB of dense matrices → 512 bytes of sign arrays - O(d²) = 16,384 ops → O(d log d) = 896 ops per rotation - Metal shader file: 1.5MB → 432KB Speed: still 2.4 tok/s. WHT reduced per-rotation cost but the bottleneck is redundant calls (8-32× per block from flash attention). The dequantize function is called per 4/16-element chunk, each time doing the full 128-element WHT. Need to modify the flash attention kernel to dequantize once per block. Quality: WHT+signs gives BETTER quality than dense QR on real KV tensors (cosine 0.94 vs 0.79 at 2-bit). Sub-Gaussian distribution (kurtosis 1.53) means fewer outliers hitting extreme centroids. Reviewed by Codex: WHT butterfly correct, inverse order verified, QJL correction matches reference C implementation. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
veniciosribeiro
added a commit
to veniciosribeiro/llama.cpp
that referenced
this pull request
Apr 8, 2026
- 10+ correções aplicadas após review - Validação de bounds em todas as funções - Correção de fórmula do produto interno assimétrico - Melhoria em tratamento de erros - Review ggml-org#2: ✅ APROVADO (task ggml-org#28)
itme-brain
pushed a commit
to itme-brain/llama.cpp
that referenced
this pull request
Apr 16, 2026
…gml-org#26 Massive reduction in constant memory and compute: - 256KB of dense matrices → 512 bytes of sign arrays - O(d²) = 16,384 ops → O(d log d) = 896 ops per rotation - Metal shader file: 1.5MB → 432KB Speed: still 2.4 tok/s. WHT reduced per-rotation cost but the bottleneck is redundant calls (8-32× per block from flash attention). The dequantize function is called per 4/16-element chunk, each time doing the full 128-element WHT. Need to modify the flash attention kernel to dequantize once per block. Quality: WHT+signs gives BETTER quality than dense QR on real KV tensors (cosine 0.94 vs 0.79 at 2-bit). Sub-Gaussian distribution (kurtosis 1.53) means fewer outliers hitting extreme centroids. Reviewed by Codex: WHT butterfly correct, inverse order verified, QJL correction matches reference C implementation. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
YuruDeveloper
pushed a commit
to YuruDeveloper/llama.cpp-quant
that referenced
this pull request
Apr 21, 2026
…gml-org#26 Massive reduction in constant memory and compute: - 256KB of dense matrices → 512 bytes of sign arrays - O(d²) = 16,384 ops → O(d log d) = 896 ops per rotation - Metal shader file: 1.5MB → 432KB Speed: still 2.4 tok/s. WHT reduced per-rotation cost but the bottleneck is redundant calls (8-32× per block from flash attention). The dequantize function is called per 4/16-element chunk, each time doing the full 128-element WHT. Need to modify the flash attention kernel to dequantize once per block. Quality: WHT+signs gives BETTER quality than dense QR on real KV tensors (cosine 0.94 vs 0.79 at 2-bit). Sub-Gaussian distribution (kurtosis 1.53) means fewer outliers hitting extreme centroids. Reviewed by Codex: WHT butterfly correct, inverse order verified, QJL correction matches reference C implementation. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Jcfunk
pushed a commit
to Jcfunk/llama.cpp
that referenced
this pull request
May 9, 2026
…gml-org#26 Massive reduction in constant memory and compute: - 256KB of dense matrices → 512 bytes of sign arrays - O(d²) = 16,384 ops → O(d log d) = 896 ops per rotation - Metal shader file: 1.5MB → 432KB Speed: still 2.4 tok/s. WHT reduced per-rotation cost but the bottleneck is redundant calls (8-32× per block from flash attention). The dequantize function is called per 4/16-element chunk, each time doing the full 128-element WHT. Need to modify the flash attention kernel to dequantize once per block. Quality: WHT+signs gives BETTER quality than dense QR on real KV tensors (cosine 0.94 vs 0.79 at 2-bit). Sub-Gaussian distribution (kurtosis 1.53) means fewer outliers hitting extreme centroids. Reviewed by Codex: WHT butterfly correct, inverse order verified, QJL correction matches reference C implementation. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
itme-brain
pushed a commit
to itme-brain/llama.cpp
that referenced
this pull request
May 11, 2026
…gml-org#26 Massive reduction in constant memory and compute: - 256KB of dense matrices → 512 bytes of sign arrays - O(d²) = 16,384 ops → O(d log d) = 896 ops per rotation - Metal shader file: 1.5MB → 432KB Speed: still 2.4 tok/s. WHT reduced per-rotation cost but the bottleneck is redundant calls (8-32× per block from flash attention). The dequantize function is called per 4/16-element chunk, each time doing the full 128-element WHT. Need to modify the flash attention kernel to dequantize once per block. Quality: WHT+signs gives BETTER quality than dense QR on real KV tensors (cosine 0.94 vs 0.79 at 2-bit). Sub-Gaussian distribution (kurtosis 1.53) means fewer outliers hitting extreme centroids. Reviewed by Codex: WHT butterfly correct, inverse order verified, QJL correction matches reference C implementation. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Jcfunk
pushed a commit
to Jcfunk/llama.cpp
that referenced
this pull request
May 13, 2026
…gml-org#26 Massive reduction in constant memory and compute: - 256KB of dense matrices → 512 bytes of sign arrays - O(d²) = 16,384 ops → O(d log d) = 896 ops per rotation - Metal shader file: 1.5MB → 432KB Speed: still 2.4 tok/s. WHT reduced per-rotation cost but the bottleneck is redundant calls (8-32× per block from flash attention). The dequantize function is called per 4/16-element chunk, each time doing the full 128-element WHT. Need to modify the flash attention kernel to dequantize once per block. Quality: WHT+signs gives BETTER quality than dense QR on real KV tensors (cosine 0.94 vs 0.79 at 2-bit). Sub-Gaussian distribution (kurtosis 1.53) means fewer outliers hitting extreme centroids. Reviewed by Codex: WHT butterfly correct, inverse order verified, QJL correction matches reference C implementation. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Jcfunk
pushed a commit
to Jcfunk/llama.cpp
that referenced
this pull request
May 13, 2026
…gml-org#26 Massive reduction in constant memory and compute: - 256KB of dense matrices → 512 bytes of sign arrays - O(d²) = 16,384 ops → O(d log d) = 896 ops per rotation - Metal shader file: 1.5MB → 432KB Speed: still 2.4 tok/s. WHT reduced per-rotation cost but the bottleneck is redundant calls (8-32× per block from flash attention). The dequantize function is called per 4/16-element chunk, each time doing the full 128-element WHT. Need to modify the flash attention kernel to dequantize once per block. Quality: WHT+signs gives BETTER quality than dense QR on real KV tensors (cosine 0.94 vs 0.79 at 2-bit). Sub-Gaussian distribution (kurtosis 1.53) means fewer outliers hitting extreme centroids. Reviewed by Codex: WHT butterfly correct, inverse order verified, QJL correction matches reference C implementation. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
meh
pushed a commit
to meh/llama.cpp
that referenced
this pull request
May 19, 2026
…-restore server: fix context checkpoint restore for hybrid/recurrent models
ckti
pushed a commit
to ckti/llama.cpp-consolidated
that referenced
this pull request
May 22, 2026
…gml-org#26 Massive reduction in constant memory and compute: - 256KB of dense matrices → 512 bytes of sign arrays - O(d²) = 16,384 ops → O(d log d) = 896 ops per rotation - Metal shader file: 1.5MB → 432KB Speed: still 2.4 tok/s. WHT reduced per-rotation cost but the bottleneck is redundant calls (8-32× per block from flash attention). The dequantize function is called per 4/16-element chunk, each time doing the full 128-element WHT. Need to modify the flash attention kernel to dequantize once per block. Quality: WHT+signs gives BETTER quality than dense QR on real KV tensors (cosine 0.94 vs 0.79 at 2-bit). Sub-Gaussian distribution (kurtosis 1.53) means fewer outliers hitting extreme centroids. Reviewed by Codex: WHT butterfly correct, inverse order verified, QJL correction matches reference C implementation. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
mndodd
added a commit
to mndodd/llama.cpp
that referenced
this pull request
Aug 12, 2026
100 commits, 8 SYCL. Conflicts in 6 files, 10 hunks. 7 union-merged (independent globals/switch-cases: our reorder_in_gemm + NVFP4 alongside their fa_onednn + Q2_K ggml-org#25064). 3 needed a call: - fattn.cpp: keep our env-gated MMA override FIRST, then upstream's oneDNN SDPA prefill FA (ggml-org#25222), then our decode VEC/TILE doors. Our decode + quantized-KV path stays ours; oneDNN takes prefill-shaped only. WARN: ggml-org#25222 has the RIG-HYGIENE ggml-org#26 pool_alloc use-after-free -- needs multi-turn 2-GPU garbage-token validation before trust. - fattn-vec.hpp: take upstream ggml-org#25205 Battlemage nthreads=256, preserve our is_vec_kernel=true arg (selects nsm geometry in launch_fattn). - dequantize.hpp q2_K: keep our f32-arithmetic dequant (NMSE 2.5e-7). test-backend-ops.cpp: v=4 was overloaded -- upstream ggml-org#25064 uses it for non-cont-a-last-2-dim, ours for transposed-b. Moved our transposed-b to bit 1<<4 so both coverages survive. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
juliantorr-es
pushed a commit
to Tribunus-dev/tessera
that referenced
this pull request
Aug 14, 2026
Pipeline refactor phases 1-4 (expert-eval seam, eval cache, KV-joint measurement plumbing, dispatch decomposition + CLI unification), the real per-tensor activation-capture feature for GA/AWQ fitness, two argument-parsing fixes (champq/w4a4 --model requirement, value-taking flags silently swallowing the next flag), and KV-joint reconstruction item 5 (scale migration: fold K/V's migrated scale into W_q/W_o). See docs/tessera-refactor-implementation-spec.md and docs/tessera-kv-joint-reconstruction-design.md for the design context. Phase gates (ggml-org#10, ggml-org#20, ggml-org#23, ggml-org#26) remain outstanding pending a real model run, not available in this environment.
Copilot AI
added a commit
to kerrrang9214-tech/llama.cpp
that referenced
this pull request
Aug 26, 2026
…L alerts ggml-org#26-ggml-org#50 Co-authored-by: kerrrang9214-tech <242015866+kerrrang9214-tech@users.noreply.github.com>
SimonTeixidor
pushed a commit
to SimonTeixidor/llama.cpp
that referenced
this pull request
Sep 13, 2026
speculative : give MTP targets their recurrent rollback slots back
tgrid0
pushed a commit
to tgrid0/llama.cpp
that referenced
this pull request
Sep 18, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #11
This fixes a Japanese prompt I was attempting to run
EG:
./main -m ./models/13B/ggml-model-q4_0.bin -t 8 -n 128 -n 512 -p $'人生の意味は'Output before change:
人生の意���、フロントカードに���いてる。 2019年3月 © All Rights Reserved. [end of text]So it is outputting some characters but some �
Output after change:
人生の意 当者： Dr. Yukari Takamatsu 作成時間: 2015年9月8日（金）、第3回ルプセンター上研修会「Mini-Workshop」で学生がしたことについて書き伝えます。 ニュアスミレショナの実行は、10位けんだあるから重要なメッセージを与り開くがうれやで報告したことについて書き伝えます。 当者: Dr. Yukari Takamatsu, MD PhD FRCR FRCP (Hon) Prof Emeritus of Hokkaido Univ School Med Sys Biol and Nanboku University Medical Sch Professor at Imperial College London Senior Member ESMO IASLC